Samsung R&D Institute China - Nanjing (SRC-N) achieved 3rd place in one of the tasks during the second edition of the Face Recognition Challenge in the Era of Synthetic Data (FRCSyn) at CVPR 2024.

The challenge is top-tier competition aiming to explore the application of synthetic data in face recognition to address the current limitations of the technology. This research line is particularly critical nowadays due to privacy concerns leading to the recent discontinuation of face recognition datasets. Furthermore, there are still shortcomings in the state-of-the-art face recognition technology related to bias in demographic groups (e.g., ethnicity and gender) and performance under challenging conditions. The results of the challenge will help in analyzing the improvements achieved while using synthetic data and the state-of-the-art performance of current face recognition technology in realistic operational scenarios, extracting significantly valuable contributions to advance in the field.

Based on our technical accumulation in the field of synthetic faces and the datasets we generated, we won third place in the Subtask 1.1.: Synthetic Data (Constrained) of Task 1: Synthetic Data for Demographic Bias Mitigation.

Because actual face datasets are not allowed, the team used exclusively synthetic data constrained to a maximum of 500,000 face images to train the face recognition model and test on specified real datasets. Our system generated 100,000 synthetic faces; the remaining 400,000 were selected from officially provided synthetic images. Improving by our generated synthetic dataset, we got the 3rd rank of Trade-Off Accuracy, 94.06 out of 100.

The SRC-Nanjing AI Software Team focuses on AIGC and intends to provide advanced AI solutions to Samsung device and services. Now, the team is engaged in the domains of visual understanding and generation and lightweight AIGC model development.